Krea-Edit-lora
Model description
license: apache-2.0 tags:
- text-to-image
- image-to-image
- lora
- diffusers
- krea-2
- character-design
- turnaround base_model: krea/krea-v2
KREA-2 Image Editing LoRA
📬 Links & Socials
- Online Workflow: Run on RunningHub
- GitHub: AIGC-Singularity
- Civitai: Civitai Profile
- Bilibili: AIGC特异点哔哩哔哩主页
- YouTube: AIGC-Singularity YouTube Channel
- WeChat: `aigctyd`
- Email: `a592991299@gmail.com`
This repository provides a specialized LoRA fine-tuned on top of KREA-2 to unlock versatile image editing and multi-view character generation, fully leveraging the outstanding aesthetic foundation of the base model.
📦 LoRA List & Models
- Model Name: `Krea2_Character_Design_4-View_V1`
- Function: Generate 4-view character turnarounds via Text-to-Image (Txt2Img) and Image-to-Image (Img2Img) workflows.
- Style Stacking: Supports loading secondary style LoRAs to generate multi-view turnaround sheets in customized artistic styles.
⚙️ Recommended Generation Settings
- Base Model: KREA-2 Turbo (`krea2_turbo`)
- LoRA Weight: `1.0`
- Sampling Steps: `8`
- CFG Scale: `1.0`
Key Features
- 4-View Character Turnaround: Supports generating 4-view character sheets (Front, Side, Back, Close-up) directly from text prompts or input reference images.
- Style LoRA Stacking: Fully compatible with secondary style LoRAs for stylized turnaround creation.
- Dual-Use Compatibility: Can be applied to both Img2Img editing workflows and standard Txt2Img generation.
- Aesthetic Preservation: Retains the rich visual style and high-end aesthetic fidelity of KREA-2.
🛠️ ComfyUI Workflow Included
The complete ComfyUI Workflow for this model has been uploaded directly to this Hugging Face repository.
You can find the workflow JSON file in the Files and versions tab (or root folder) of this repo. Download it and drop it into your ComfyUI interface to quickly set up both Txt2Img and Img2Img multi-view generation pipelines!
Usage Example (`diffusers`)
```python import torch from diffusers import AutoPipelineForImage2Image
Load KREA-2 Turbo base model
pipe = AutoPipelineForImage2Image.from_pretrained( "krea/krea-v2-turbo", torch_dtype=torch.float16 ).to("cuda")
Load 4-View LoRA with weight 1.0
pipe.load_lora_weights("your-username/krea2-editing-lora", weight_name="Krea2_Character_Design_4-View_V1.safetensors")
Run 4-View Turnaround Generation with recommended settings (Steps: 8, CFG: 1.0)
prompt = "4-view turnaround of a character, front view, side view, back view, facial close-up, plain solid white background" output_image = pipe( prompt=prompt, image=init_image, num_inference_steps=8, guidance_scale=1.0, cross_attention_kwargs={"scale": 1.0} ).images[0]
output_image.save("4view_output.png")
Download model
Download them in the Files & versions tab.